Añade funciones de previsualización de texto y manejo de historial en `app.py`. Se implementan las funciones `_preview_text` y `_history_preview` para limitar la longitud del texto mostrado y mejorar la visualización del historial de interacciones. Además, se integra la instrumentación opcional de RunTree para el seguimiento de sesiones de chat, permitiendo registrar la cantidad de mensajes y manejar errores de manera más efectiva.

app.py

@@ -5,6 +5,7 @@ import gradio as gr
 from openai import OpenAI
 from dotenv import load_dotenv
 from langsmith import traceable
+from langsmith.run_trees import RunTree
 
 load_dotenv()
 
@@ -91,6 +92,28 @@ def _extract_text_and_files(message):
     return text_combined, files
 
 
+def _preview_text(text: str | None, limit: int = 600) -> str:
+    if not text:
+        return ""
+    if len(text) <= limit:
+        return text
+    return text[:limit] + "…"
+
+
+def _history_preview(history: list[tuple[str, str]] | None, max_turns: int = 3, max_chars: int = 1200) -> str:
+    if not history:
+        return ""
+    tail = history[-max_turns:]
+    parts: list[str] = []
+    for user_turn, assistant_turn in tail:
+        if user_turn:
+            parts.append(f"U: {_preview_text(user_turn, 300)}")
+        if assistant_turn:
+            parts.append(f"A: {_preview_text(assistant_turn, 300)}")
+    joined = "\n".join(parts)
+    return _preview_text(joined, max_chars)
+
+
 @traceable
 def respond(message, history: list[tuple[str, str]]):
     """Stream assistant reply via Gemini using OpenAI-compatible API.
@@ -153,7 +176,59 @@ def respond(message, history: list[tuple[str, str]]):
 
     messages.append({"role": "user", "content": user_content})
 
+    # Optional RunTree instrumentation (does not require LANGSMITH_TRACING)
+    _ls_api_key = os.getenv("LANGSMITH_API_KEY")
+    pipeline = None
+    child_build = None
+    child_llm = None
+    if _ls_api_key:
+        try:
+            pipeline = RunTree(
+                name="Chat Session",
+                run_type="chain",
+                inputs={
+                    "user_text": _preview_text(user_text, 600),
+                    "has_images": bool(image_parts),
+                    "history_preview": _history_preview(history),
+                },
+            )
+            pipeline.post()
+
+            child_build = pipeline.create_child(
+                name="BuildMessages",
+                run_type="chain",
+                inputs={
+                    "system_prompt_preview": _preview_text(system_prompt, 400),
+                    "user_content_type": "multimodal" if image_parts else "text",
+                    "history_turns": len(history or []),
+                },
+            )
+            child_build.post()
+            child_build.end(
+                outputs={
+                    "messages_count": len(messages),
+                }
+            )
+            child_build.patch()
+        except Exception:
+            pipeline = None
+
     try:
+        if pipeline:
+            try:
+                child_llm = pipeline.create_child(
+                    name="LLMCall",
+                    run_type="llm",
+                    inputs={
+                        "model": GEMINI_MODEL,
+                        "provider": "gemini-openai",
+                        "messages_preview": _preview_text(str(messages[-1]), 600),
+                    },
+                )
+                child_llm.post()
+            except Exception:
+                child_llm = None
+
         stream = _client.chat.completions.create(
             model=GEMINI_MODEL,
             messages=messages,
@@ -180,7 +255,32 @@ def respond(message, history: list[tuple[str, str]]):
 
         if not accumulated:
             yield "(Sin contenido de respuesta)"
+
+        if child_llm:
+            try:
+                child_llm.end(outputs={"content": _preview_text(accumulated, 5000)})
+                child_llm.patch()
+            except Exception:
+                pass
+        if pipeline:
+            try:
+                pipeline.end(outputs={"answer": _preview_text(accumulated, 5000)})
+                pipeline.patch()
+            except Exception:
+                pass
     except Exception as e:
+        if child_llm:
+            try:
+                child_llm.end(outputs={"error": str(e)})
+                child_llm.patch()
+            except Exception:
+                pass
+        if pipeline:
+            try:
+                pipeline.end(outputs={"error": str(e)})
+                pipeline.patch()
+            except Exception:
+                pass
         yield f"Ocurrió un error al llamar a Gemini: {e}"